1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/sysproto.h>
41 #include <sys/kernel.h>
42 #include <sys/sysctl.h>
43 #include <sys/file.h>
44 #include <sys/filedesc.h>
45 #include <sys/vnode.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/proc.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/mbuf.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/domain.h>
55 #include <sys/protosw.h>
56 #include <sys/namei.h>
57 #include <sys/unistd.h>
58 #include <sys/kthread.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
61 #include <sys/mutex.h>
62 #include <sys/taskqueue.h>
63
64 #include <netinet/in.h>
65 #include <netinet/tcp.h>
66
67 #include <rpc/rpcclnt.h>
68
69 #include <nfs/xdr_subs.h>
70 #include <nfs/rpcv2.h>
71 #include <nfs/nfsproto.h>
72 #include <nfsclient/nfs.h>
73 #include <nfsclient/nfsm_subs.h>
74 #include <nfsclient/nfsmount.h>
75 #include <nfsclient/nfsnode.h>
76 #include <nfsclient/nfs_lock.h>
77
78 static MALLOC_DEFINE(M_NFSSVC, "nfsclient_srvsock", "Nfs server structure");
79
80 static void nfssvc_iod(void *);
81
82 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
83
84 SYSCTL_DECL(_vfs_nfs);
85
86 /* Maximum number of seconds a nfsiod kthread will sleep before exiting */
87 static unsigned int nfs_iodmaxidle = 120;
88 SYSCTL_UINT(_vfs_nfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0, "");
89
90 /* Maximum number of nfsiod kthreads */
91 unsigned int nfs_iodmax = 20;
92
93 /* Minimum number of nfsiod kthreads to keep as spares */
94 static unsigned int nfs_iodmin = 0;
95
96 static int nfs_nfsiodnew_sync(void);
97
98 static int
99 sysctl_iodmin(SYSCTL_HANDLER_ARGS)
100 {
101 int error, i;
102 int newmin;
103
104 newmin = nfs_iodmin;
105 error = sysctl_handle_int(oidp, &newmin, 0, req);
106 if (error || (req->newptr == NULL))
107 return (error);
108 mtx_lock(&nfs_iod_mtx);
109 if (newmin > nfs_iodmax) {
110 error = EINVAL;
111 goto out;
112 }
113 nfs_iodmin = newmin;
114 if (nfs_numasync >= nfs_iodmin)
115 goto out;
116 /*
117 * If the current number of nfsiod is lower
118 * than the new minimum, create some more.
119 */
120 for (i = nfs_iodmin - nfs_numasync; i > 0; i--)
121 nfs_nfsiodnew_sync();
122 out:
123 mtx_unlock(&nfs_iod_mtx);
124 return (0);
125 }
126 SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0,
127 sizeof (nfs_iodmin), sysctl_iodmin, "IU", "");
128
129
130 static int
131 sysctl_iodmax(SYSCTL_HANDLER_ARGS)
132 {
133 int error, i;
134 int iod, newmax;
135
136 newmax = nfs_iodmax;
137 error = sysctl_handle_int(oidp, &newmax, 0, req);
138 if (error || (req->newptr == NULL))
139 return (error);
140 if (newmax > NFS_MAXASYNCDAEMON)
141 return (EINVAL);
142 mtx_lock(&nfs_iod_mtx);
143 nfs_iodmax = newmax;
144 if (nfs_numasync <= nfs_iodmax)
145 goto out;
146 /*
147 * If there are some asleep nfsiods that should
148 * exit, wakeup() them so that they check nfs_iodmax
149 * and exit. Those who are active will exit as
150 * soon as they finish I/O.
151 */
152 iod = nfs_numasync - 1;
153 for (i = 0; i < nfs_numasync - nfs_iodmax; i++) {
154 if (nfs_iodwant[iod] == NFSIOD_AVAILABLE)
155 wakeup(&nfs_iodwant[iod]);
156 iod--;
157 }
158 out:
159 mtx_unlock(&nfs_iod_mtx);
160 return (0);
161 }
162 SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0,
163 sizeof (nfs_iodmax), sysctl_iodmax, "IU", "");
164
165 static int
166 nfs_nfsiodnew_sync(void)
167 {
168 int error, i;
169
170 mtx_assert(&nfs_iod_mtx, MA_OWNED);
171 for (i = 0; i < nfs_iodmax; i++) {
172 if (nfs_asyncdaemon[i] == 0) {
173 nfs_asyncdaemon[i] = 1;
174 break;
175 }
176 }
177 if (i == nfs_iodmax)
178 return (0);
179 mtx_unlock(&nfs_iod_mtx);
180 error = kthread_create(nfssvc_iod, nfs_asyncdaemon + i, NULL,
181 RFHIGHPID, 0, "nfsiod %d", i);
182 mtx_lock(&nfs_iod_mtx);
183 if (error == 0) {
184 nfs_numasync++;
185 nfs_iodwant[i] = NFSIOD_AVAILABLE;
186 } else
187 nfs_asyncdaemon[i] = 0;
188 return (error);
189 }
190
191 void
192 nfs_nfsiodnew_tq(__unused void *arg, int pending)
193 {
194
195 mtx_lock(&nfs_iod_mtx);
196 while (pending > 0) {
197 pending--;
198 nfs_nfsiodnew_sync();
199 }
200 mtx_unlock(&nfs_iod_mtx);
201 }
202
203 void
204 nfs_nfsiodnew(void)
205 {
206
207 mtx_assert(&nfs_iod_mtx, MA_OWNED);
208 taskqueue_enqueue(taskqueue_thread, &nfs_nfsiodnew_task);
209 }
210
211 static void
212 nfsiod_setup(void *dummy)
213 {
214 int error;
215
216 TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin);
217 mtx_lock(&nfs_iod_mtx);
218 /* Silently limit the start number of nfsiod's */
219 if (nfs_iodmin > NFS_MAXASYNCDAEMON)
220 nfs_iodmin = NFS_MAXASYNCDAEMON;
221
222 while (nfs_numasync < nfs_iodmin) {
223 error = nfs_nfsiodnew_sync();
224 if (error == -1)
225 panic("nfsiod_setup: nfs_nfsiodnew failed");
226 }
227 mtx_unlock(&nfs_iod_mtx);
228 }
229 SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL);
230
231 static int nfs_defect = 0;
232 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
233
234 /*
235 * Asynchronous I/O daemons for client nfs.
236 * They do read-ahead and write-behind operations on the block I/O cache.
237 * Returns if we hit the timeout defined by the iodmaxidle sysctl.
238 */
239 static void
240 nfssvc_iod(void *instance)
241 {
242 struct buf *bp;
243 struct nfsmount *nmp;
244 int myiod, timo;
245 int error = 0;
246
247 mtx_lock(&nfs_iod_mtx);
248 myiod = (int *)instance - nfs_asyncdaemon;
249 /*
250 * Main loop
251 */
252 for (;;) {
253 while (((nmp = nfs_iodmount[myiod]) == NULL)
254 || !TAILQ_FIRST(&nmp->nm_bufq)) {
255 if (myiod >= nfs_iodmax)
256 goto finish;
257 if (nmp)
258 nmp->nm_bufqiods--;
259 if (nfs_iodwant[myiod] == NFSIOD_NOT_AVAILABLE)
260 nfs_iodwant[myiod] = NFSIOD_AVAILABLE;
261 nfs_iodmount[myiod] = NULL;
262 /*
263 * Always keep at least nfs_iodmin kthreads.
264 */
265 timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz;
266 error = msleep(&nfs_iodwant[myiod], &nfs_iod_mtx, PWAIT | PCATCH,
267 "-", timo);
268 if (error) {
269 nmp = nfs_iodmount[myiod];
270 /*
271 * Rechecking the nm_bufq closes a rare race where the
272 * nfsiod is woken up at the exact time the idle timeout
273 * fires
274 */
275 if (nmp && TAILQ_FIRST(&nmp->nm_bufq))
276 error = 0;
277 break;
278 }
279 }
280 if (error)
281 break;
282 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
283 int giant_locked = 0;
284
285 /* Take one off the front of the list */
286 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
287 nmp->nm_bufqlen--;
288 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) {
289 nmp->nm_bufqwant = 0;
290 wakeup(&nmp->nm_bufq);
291 }
292 mtx_unlock(&nfs_iod_mtx);
293 if (NFS_ISV4(bp->b_vp)) {
294 giant_locked = 1;
295 mtx_lock(&Giant);
296 }
297 if (bp->b_flags & B_DIRECT) {
298 KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set"));
299 (void)nfs_doio_directwrite(bp);
300 } else {
301 if (bp->b_iocmd == BIO_READ)
302 (void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL);
303 else
304 (void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL);
305 }
306 if (giant_locked)
307 mtx_unlock(&Giant);
308 mtx_lock(&nfs_iod_mtx);
309 /*
310 * If there are more than one iod on this mount, then defect
311 * so that the iods can be shared out fairly between the mounts
312 */
313 if (nfs_defect && nmp->nm_bufqiods > 1) {
314 NFS_DPF(ASYNCIO,
315 ("nfssvc_iod: iod %d defecting from mount %p\n",
316 myiod, nmp));
317 nfs_iodmount[myiod] = NULL;
318 nmp->nm_bufqiods--;
319 break;
320 }
321 }
322 }
323 finish:
324 nfs_asyncdaemon[myiod] = 0;
325 if (nmp)
326 nmp->nm_bufqiods--;
327 nfs_iodwant[myiod] = NFSIOD_NOT_AVAILABLE;
328 nfs_iodmount[myiod] = NULL;
329 /* Someone may be waiting for the last nfsiod to terminate. */
330 if (--nfs_numasync == 0)
331 wakeup(&nfs_numasync);
332 mtx_unlock(&nfs_iod_mtx);
333 if ((error == 0) || (error == EWOULDBLOCK))
334 kthread_exit(0);
335 /* Abnormal termination */
336 kthread_exit(1);
337 }
Cache object: 4bdb10d3c7d393da40309a6e247eb145
|